Genetic algorithm based simultaneous optimization of feature subsets and hidden Markov model parameters for discrimination between speech and non-speech events

Li, Yanxiong; He, Qianhua; He, Jun; Yang, Jichen

doi:10.1007/s10772-010-9070-4

Cited by 10 publications

(7 citation statements)

References 24 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…GA is an optimization algorithm that simulates Darwin's genetic choice and biological natural selection process. (38,39) It uses a coding scheme to realize parameter optimization through iterative genetic operations, including selection, crossover, and mutation, where a fitness function is usually defined to find the optimal solution. For HMM parameter initialization in this study, floating point number coding is used as the coding scheme due to the high precision requirement when the mixed Gaussian probability density function is employed for generating the observation sequence.…”

Section: Ga Initializationmentioning

confidence: 99%

Analyzing Wrist Pulse Signals Measured with Polyvinylidene Fluoride Film for Hypertension Identification

2017

Sensors and Materials

View full text Add to dashboard Cite

The human wrist pulse signal, which often exhibits nonstationarity, can reflect changes in mechanisms and pathophysiology in the blood and viscera. In this paper, we present an integrated approach for identifying hypertension using the wavelet packet transform (WPT) and continuous hidden Markov models (HMM) to analyze wrist pulse signals. The approach starts with decomposing the wrist pulse signals into a number of frequency sub-bands through the WPT. Then the local discriminant bases (LDB) algorithm is used to obtain the best representation of the wrist pulse signal in the optimal frequency sub-bands. After that, energy features are extracted from those sub-bands and the optimal features associated with corresponding sub-bands are selected using the Fisher linear discriminant criterion. The optimal features are subsequently used as input to a continuous HMM classifier for hypertension identification. Experimental results indicate that the presented approach can differentiate the hypertensive wrist pulses from healthy wrist pulses effectively. In addition, when compared with other classifiers, the results demonstrate that the continuous HMM classifier yields a better classification result.

show abstract

Section: Ga Initializationmentioning

confidence: 99%

Analyzing Wrist Pulse Signals Measured with Polyvinylidene Fluoride Film for Hypertension Identification

2017

Sensors and Materials

View full text Add to dashboard Cite

show abstract

“…For example, a methodology for learning specialised filter banks using deep neural networks was proposed in [6]. Moreover, several approaches based on evolutionary computation have been proposed for the search of optimal speech representations [7][8][9][10].…”

Section: Introductionmentioning

confidence: 99%

Multi‐objective optimisation of wavelet features for phoneme recognition

Vignolo

Rufiner

Milone

2016

IET signal process.

View full text Add to dashboard Cite

One of the most important issues in speech applications involves the preprocessing stage, which is meant to produce a manageable set of significant features, exploiting the capabilities of the classification phase [5]. The most widely used features for speech recognition, and also applied for different tasks involving speech and music signals, are the mel-frequency cepstral coefficients (MFCCs) [1,5]. These are based on the linear model of voice production and a psychoacoustic scale [5]. Even though MFCCs provide acceptable performance under laboratory conditions, recognition rates degrade significantly in presence of noise. This has motivated many advances in the development of robust feature extraction approaches, like perceptual linear prediction (PLP) and relative spectra [1]. More recently, speech processing techniques based on computational intelligence tools have been developed. For example, several approaches based on evolutionary computation have been proposed for the search of optimal speech representations [8]. Wavelet based processing provides useful tools for the analysis of nonstationary signals, which have been found suitable for speech feature extraction [6]. In order to build a representation based on the wavelet packet transform (WPT), frequently a particular orthogonal basis is selected among all the available basis [6]. However, for speech recognition there is no evidence showing the convenience of the use of orthogonal basis. Therefore, removing the orthogonality restriction the complete WPT decomposition offers a highly redundant set of coefficients, some of which can be selected to build an optimal representation.The optimisation of wavelet decompositions for feature extraction has been studied in many different ways, though it is still an open challenge in speech processing. For example, the optimisation of wavelet decompositions by means of evolutionary algorithms was proposed for image watermarking [4] and for signal denoising [2]. In [9] we proposed a novel approach for the optimisation of over-complete decompositions from a WPT dictionary based on a multi-objective genetic algorithm (MOGA). The MOGA allows to maximise the classification accuracy while minimising the number of features. For the purpose of obtaining appropriate features for state of the art speech recognizers, a classifier based on hidden Markov models (HMM) is used to estimate the capability of candidate solutions, using on a set of English phonemes. The proposed method, which we refer to as evolutionary wavelet packets (EWP), exploits the benefits provided

show abstract

“…On the other hand, significant progress has been made with the application of different artificial intelligence techniques for feature selection. In particular, many works rely on evolutionary algorithms for feature subset optimization [10], [11], and for the search of optimal representations [12], [13], [14].…”

Section: Introductionmentioning

confidence: 99%

An evolutionary wrapper for feature selection in face recognition applications

Vignolo

Milone

Behaine

et al. 2012

2012 IEEE International Conference on Systems, Man, and Cybernetics (SMC)

View full text Add to dashboard Cite

Active shape models is an adaptive shape-matching technique that has been used for locating facial features in images. However, when a number of features is extracted for each landmark point, distortions caused by noise or illumination, and the dimensionality of the final representation, have a negative impact in the performance of a classifier. In this paper, an evolutionary wrapper for selection of the most relevant set of features for face recognition is presented. The proposed strategy explores the space of multiple feasible selections using genetic algorithms. Experimental results show that the proposed approach allows to improve the classification performance in comparison with another enhanced method and a state of the art face recognition approach.

show abstract

Genetic algorithm based simultaneous optimization of feature subsets and hidden Markov model parameters for discrimination between speech and non-speech events

Cited by 10 publications

References 24 publications

Analyzing Wrist Pulse Signals Measured with Polyvinylidene Fluoride Film for Hypertension Identification

Analyzing Wrist Pulse Signals Measured with Polyvinylidene Fluoride Film for Hypertension Identification

Multi‐objective optimisation of wavelet features for phoneme recognition

An evolutionary wrapper for feature selection in face recognition applications

Contact Info

Product

Resources

About